Remnant44

19 ms·

31.
by Remnant44 1y ago
I mean... A single 9950x core is going to struggle to do more than 16 GB/second of direct mem copy bandwidth. So being within an order of magnitude of that seems reasonable
32.
by Remnant44 1y ago
You're missing half the equation. It's Risk vs Reward. Anyone who built something great took huge risks, for a huge return. There is no return here. At best, you've sunk a ton of capital into a low-profit business that is pro
33.
by Remnant44 1y ago
It doesn't make it go away. Using your analogy - if you used 300lb to learn and then only made 10 lb of pizza after that, it would be a pretty poor use of resources. If you instead went on to produce millions of pizzas for people and 3
34.
by Remnant44 1y ago
I don't really follow this objection. Determining the actual energy usage of AI training+inference is something that is an objective reality. Whether you hate it or love it doesn't change these facts.
35.
by Remnant44 1y ago
By the same token, even if you accept that AI usage will incentivize additonal model trainings, that cost is diffused across hundreds of millions of users, and is not a marginal cost, so it gets further reduced on a per-chat basis the more
36.
by Remnant44 1y ago
That's a good point about increased dependency chain length in simd due to the branchless programming style. Unrolling to break a loop-carried dependency is one of the strongest reasons to unroll especially simd code. Unrolling trivial
37.
by Remnant44 1y ago
My understanding is that they've changed this for Lion Cove and all future P cores, moving to much more of a Zen-like setup with seperate schedulers and ports for vector and scalar ops.
38.
by Remnant44 1y ago
These days, I strongly believe that loop unrolling is a pessimization, especially with SIMD code. Scalar code should be unrolled by the compiler to the SIMD word width to expose potential parallelism. But other than that, correctly predicte
39.
by Remnant44 2y ago
Thunderbird is my daily driver email client for all my business email for over a decade. I've never had a single corruption problem in that time, with probably hundreds of thousands of emails. Take that for what its worth. The only com
40.
by Remnant44 2y ago
Well. That's a pleasant surprise - Intel making the right move here! There was so much unforced error around avx512 strategy. According to the steam hardware survey for Feb 2025, AVX512 support is already up to around ~16%. That may no
41.
by Remnant44 2y ago
Another one I learned recently: The godot engine is not go-dot, it's guh-dow.
42.
by Remnant44 2y ago
Nothing to fix here. The behavior in the transition regimes is already quite good. The overall throttling is dynamic and reactive based on heat and power draw - this is unavoidable and in fact desirable (the alternative is to simply run slo
43.
by Remnant44 2y ago
Appreciate the detail! That explains a lot of what is going on.. It also dovetails with some interesting facts I remember reading about the relative power consumption for the zen cores versus the infinity fabric connecting them - The percen
44.
by Remnant44 2y ago
It's interesting that Zen5's FPUs running in full 512bit wide mode doesn't actually seem to cause any trouble, but that lighting up the load store units does. I don't know enough about hardware-level design to know if th
45.
by Remnant44 2y ago
I thought this was a well written and fair article. My preference would be that C++ is not "forked", but I understand the reasons for the impasse and consequences. I found the part about member access operators interesting. I&#x27
46.
by Remnant44 2y ago
This is my favorite passage from the article: "Beyond this, people make mistakes. Brilliance represents an upper bound on the quality of your reasoning, but there is no lower bound. The most brilliant scientist in the world can take re
47.
by Remnant44 2y ago
> An orange is 100% fructose It is not: https://fdc.nal.usda.gov/food-details/746771/nutrients In fact an orange is a mixture of glucose, fructose, and sucrose (which itself is glucose and fructose). An orange
48.
by Remnant44 2y ago
My understanding is that it's a pretty small optimization on the front end. It doesn't have a lot of entries to begin with (144) so the amount of space saved is probably negligible. Theoretically, the loop buffer would let you sav
49.
by Remnant44 2y ago
I would wager that most real world SIMD use is with direct intrinsics.
50.
by Remnant44 2y ago
utf8, for essentially the reasons mentioned in this manifesto: https://utf8everywhere.org/
51.
by Remnant44 2y ago
Agreed.. in a lot of ways, APX is like AVX(512) for scalar code. Double integer register count, three-operand instructions, improved predication support including conditional load/stores.. If I remember correctly, they actually impleme
52.
by Remnant44 2y ago
This is hopeful. Whatever the ARM enthusiasts would like, x86 is going to stick around for a long time, and working together to evolve the ISA extensions in a more cohesive manner would go a long way. In particular, I'd really like AMD
53.
by Remnant44 2y ago
I agree! I think people are used to comparing to a single threaded execution of non-vectorized code, which is using .1% of a modern CPU's compute power. Where the balance slants all the way towards gpus again is the tensor units using
54.
by Remnant44 2y ago
The execution units are fully pipelined, so although the latency is four cycles, you can receive one result every cycle from each of the execution units. For a Zen 5 core, that means 16 double precision FMAs per cycle using AVX 512, so 80gf
55.
by Remnant44 2y ago
First of all, I love this concept! The editable fields within a markdown explainer is really intuitive. What I'd personally like to see is a better data exploration tab or similar - basically some place that makes it easier to view the
56.
by Remnant44 2y ago
As mentioned in other responses, that part of the GPU simply isn't interesting from an architectural perspective, which is what Chips and Cheese is all about. GPU compute performance is both technically interesting, and matters to much
57.
by Remnant44 2y ago
This is the right answer. :) I think a lot of folks get mixed up with GPU and/or SIMD architecture, where you execute both sides of the branch out of necessity: Some of the lanes need to go one way and some the other, so you have to do
58.
by Remnant44 2y ago
Which is a very important thing to amplify. Healthcare in the US is full of gotchas and small exceptions that can be absolutely devastating if you fall through the cracks. But Obamacare's mandate of coverage for pre-existing conditions
59.
by Remnant44 2y ago
It's basically like clockwork - you can assume that any post about c++ language evolution is going to have a number of people saying one or all of the following : 1) "CPP keeps getting complex in useless ways. Just use C (maybe C
60.
by Remnant44 2y ago
I haven't had a high DPI issue with QT in quite a while. Early in QT 5, high DPI presented many problems, but they have been slowly settled over time, including by using sane defaults. Source: I develop a QT widget app and develop on a

More ›